up

Author: sayakpaul

src/diffusers/models/attention_dispatch.py

@@ -369,7 +369,7 @@ def _check_attention_backend_requirements(backend: AttentionBackendName) -> None
     elif backend in [AttentionBackendName._FLASH_3_HUB]:
         if not DIFFUSERS_ENABLE_HUB_KERNELS:
             raise RuntimeError(
-                f"Flash Attention 3 Hub backend '{backend.value}' is not usable because the `DIFFUSERS_ENABLE_HUB_KERNELS` env var isn't set. Please set it `DIFFUSERS_ENABLE_HUB_KERNELS=yes`."
+                f"Flash Attention 3 Hub backend '{backend.value}' is not usable because the `DIFFUSERS_ENABLE_HUB_KERNELS` env var isn't set. Please set it like `export DIFFUSERS_ENABLE_HUB_KERNELS=yes`."
             )
         if not is_kernels_available():
             raise RuntimeError(

src/diffusers/utils/kernels_utils.py

@@ -16,7 +16,7 @@ def _get_fa3_from_hub():
 
         try:
             # TODO: temporary revision for now. Remove when merged upstream into `main`.
-            flash_attn_3_hub = get_kernel(_DEFAULT_HUB_ID_FA3, revision="fake-ops-return-prob")
+            flash_attn_3_hub = get_kernel(_DEFAULT_HUB_ID_FA3, revision="fake-ops-return-probs")
             return flash_attn_3_hub
         except Exception as e:
             logger.error(f"An error occurred while fetching kernel '{_DEFAULT_HUB_ID_FA3}' from the Hub: {e}")